Analysis of the manufacturing signature using data mining

Abstract

The use of data mining within manufacturing is a relatively modern application. Data mining can be used to find underlying links between the machining conditions, and parameters, and the final form of the part. Part of this procedure includes defining the form of the part, known as the manufacturing signature, which stems from all steps in the manufacturing process. In this paper, two potential definitions for the manufacturing signature of cylindrical objects are generated in terms of an analytical basis. The first description uses a simple Fourier description (known as lobing) and the second consists of a fully orthonormal description in terms of Forsythe polynomials and Fourier coefficients. Principal Component Analysis (PCA) is also partially used to investigate the underlying structure of the cylinders and investigate the connection between the analytical description and PCA. Experiments were carried out, machining thirty components under different manufacturing conditions (such as coolant pressure, tool length etc.). Data mining was then carried out on the process parameters, and either the amount of a given type of lobing or the classification of the cylinder in terms of the maximal lobing. The input to data mining for our case is either a numeric answer or a classification, which motivates the use of a simplified description. The use of PCA on this data set indicates a fundamental issue stemming from subsets of “similar” data which means dimensionality reduction is not possible in the usual way. The use of the analytical basis suggests a new sampling strategy to be used on certain geometries utilising Gauss-Legendre quadrature.